Predictive modeling for intelligent maintenance in complex semiconductor manufacturing processes

Semiconductor fabrication is one of the most complicated manufacturing processes, in which the current prevailing maintenance practices are preventive maintenance, using either time-based or wafer-based scheduling strategies, which may lead to the tools being either "over-maintained" or "under-maintained". In literature, there rarely exists condition-based maintenance, which utilizes machine conditions to schedule maintenance, and almost no truly predictive maintenance that assesses remaining useful lives of machines and plans maintenance actions proactively.; The research presented in this thesis is aimed at developing predictive modeling methods for intelligent maintenance in semiconductor manufacturing processes, using the in-process tool performance as well as the product quality information. In order to achieve an improved maintenance decision-making, a method for integrating data from different domains to predict process yield is proposed. The self-organizing maps have been utilized to discretize continuous data into discrete values, which will tremendously reduce the computational cost of Bayesian network learning process that can discover the stochastic dependences among process parameters and product quality. This method enables one to make more proactive product quality prediction that is different from traditional methods based on solely inspection results.; Furthermore, a method of using observable process information to estimate stratified tool degradation levels has been proposed. Single hidden Markov model (HMM) has been employed to represent the tool degradation process under a single recipe; and the concatenation of multiple HMMs can be used to model the tool degradation under multiple recipes. To validate the proposed method, a simulation study has been conducted, which shows that HMMs are able to model the stratified unobservable degradation process under variable operating conditions. This method enables one to estimate the condition of in-chamber particle contamination so that maintenance actions can be initiated accordingly.; With these two novel methods, a methodological framework to perform better maintenance in complex manufacturing processes is established. The simulation study shows that the maintenance cost can be reduced by performing predictive maintenance properly while highest possible yield is retained. This framework provides a possibility of using abundant equipment monitoring data and product quality information to coordinate maintenance actions in a complex manufacturing environment.